Units
1. Introduction to Artificial Intelligence in Medicine: Boundaries, Validation, Responsibility and Ethics 2. Clinical Decision Support: Differential Diagnosis Recommendation and Physician Approval 3. Medical Literature Review and Evidence Summary 4. Writing Patient History (Anamnesis) and Epicrisis 5. Imaging Pre-Assessment and Radiology AI 6. Drug Interaction, Dosage and Prescription Safety Control 7. Patient Information, Health Literacy and Communication 8. Hallucination, Reliability, and Evidence-Based Validation 9. KVKK, Patient Privacy and Data Security 10. Clinical Documentation, Coding and Administrative Workflow 11. Responsibility, Law, Ethics and Informed Consent 12. End-to-End Artificial Intelligence Supported Clinical Workflow and Enterprise Application
Unit 4 / 12

Writing Patient History (Anamnesis) and Epicrisis

Gains:

  • Ability to produce anamnesis, epicrisis and clinical note draft quickly and consistently with structured prompts
  • Ability to check fabricated findings, wrong dates and inconsistencies in the text produced by artificial intelligence from a physician's perspective
  • Ability to protect patient privacy and exact accuracy with the real file in clinical texts

Anamnesis is the patient's history: when and how the complaint started; past illnesses, medications, habits. Epicrisis is a summary report of a hospitalization or treatment process; It describes how the patient arrived, how he was treated, and how he was discharged. These documents are the basis of clinical care and the legal record. Writing takes time and takes the physician away from the bedside. This is where artificial intelligence (AI) is valuable: turning messy notes into organized text, reducing repetition, providing consistent language. But there is one critical risk: the AI ​​may add findings that are not in your notes “just because they seem plausible” (hallucination). A clinical document is a legal record; Each line must be exactly correct with the actual file. In this unit, you will learn to use AI safely in anamnesis and epicrisis writing.

The power and pitfalls of AI in text generation

The power of AI is to turn messy input into organized output. It turns short notes such as "62 years old male, 3 days of chest pain, HT, ECG normal, troponin negative, observation" into a well-titled and fluent epicrisis. This saves minutes and improves readability.

The trap is the AI's tendency to "complement". The model has learned what is found in a typical epicrisis; Therefore, even if you do not give it, they may add statements such as "no pathology was detected in the physical examination" or "fever is 38 degrees". If these are not in the actual file, the document is incorrect and legally problematic. Additionally, AI can confuse dates, doses, and lab values. So the golden rule: let AI only edit the information you provide; Do not add any findings.

Attention: Epicrisis and anamnesis are legal documents. Every line the AI ​​produces cannot be signed without comparing it to the actual file. A fabricated finding (non-existent fever, non-existent allergy) poses both clinical and legal risks.

Step by step: Safe clinical text with AI

  1. Collect notes anonymously and completely. Prepare all clinical data without identifying information.
  2. Give strict rule to AI. "Just edit the information I provided; do not add any findings, values, or dates. If there are any missing fields, write [MISSING]."
  3. Specify the structure. Headings: anamnesis, examination, examination, diagnosis, treatment, recommendation.
  4. Compare the output to the file line by line. Does every finding really exist?
  5. Delete anything added/fabricated. Fill in [MISSING] fields with actual data.
  6. As a physician, read, correct and sign. The responsibility is on you.

three mini cases

Case 1 — Fake finding caught. A physician notices that the AI ​​added "fever 38.5 degrees" to the epicrisis when there was no fever information in his notes. In the real file, the temperature is 36.8. The physician deletes this line. If it had not been checked, an incorrect vital sign would have been entered in the official document.

Case 2 — Date confusion. In the epicrisis of a 4-day hospitalization, AI writes the admission and discharge dates backwards and puts an examination on the wrong day. The doctor compares the date order with the file and corrects it. The wrong date could disrupt both clinical flow and reimbursement.

Case 3 — Correct and fast use. During a busy outpatient day, the physician anonymously converts 12 patients' short notes into regular epicrises with AI; It compares each one with the file and makes an average of 3 corrections. Total writing time decreases by more than half, and language consistency increases. AI wrote, physician confirmed and signed.

AI output audit chart

Will be checked

risky behavior

correct behavior

Findings

A non-existent finding was added

Only the findings in the file

vital/laboratory

Value made up/mixed up

Exactly with real values

Dates

Admission/discharge reversed

Chronology same as file

Drug/dose

Dose added/changed

Exactly on prescription

ID

Name/TC is left in the text

Anonymous, KVKK compliant

Four copyable templates

Role: You are a clinical copy editor. You are not a doctor; YOU CANNOT ADD information.Task: Turn the following loose notes into an organized EPICRISE outline.RULE: Only use the information I give. DO NOT ADD any signs, vitals, laboratory, date or dose. Mark the missing field as [MISSING].Titles: Anamnesis / Physical Examination / Examinations / Diagnosis / Treatment / Recommendations.Notes (anonymous): [...]

Task: Compare the draft of the epicrisis below with the ORIGINAL notes I gave. Mark every sentence, value and date that is NOT in the notes. Give a separate list of anything that might have been made up/added. Original notes: [...]Draft: [...]

Task: Organize the following patient history according to systematic anamnesis headings: complaint, history, CV, family history, medications, habits, allergies. Add information; leave missing title [MISSING].Raw story (anonymous): [...]

Task: Identify any identification information (name, surname, TR ID, file number, telephone, address) that may remain in the text below and replace it with [ANONOMY]. Do not distort the clinical meaning. Text: [...]

Weak prompt / Strong prompt

Weak: "Write a nice epicrisis for this patient."

Güçlü: "Turn the anonymous notes below into an organized epicrisis. Use only the information I provided; do not add any findings, vitals, laboratory values, date or dose. Leave the missing fields as [MISSING]. Headings: Anamnesis, Physical Examination, Examinations, Diagnosis, Treatment, Recommendations. At the end, if there is something you added that is not in the notes, write it down as a separate list."

In the powerful prompt, AI's door to adding information is closed and control is simplified.

Common mistakes

  • Not noticing the fabricated finding. Can add non-AI fever/allergy/exam.
  • Not checking dates and values. AI can confuse chronology and laboratory.
  • Leaving identification information in the text. Risk of KVKK violation.
  • Signing the printout without reading it. The responsibility lies with the physician; Every row must be validated.
  • Not providing the "add information" rule. Without rules, AI fills in the gaps.

Voice memo and speech-to-text attention

An increasingly common use is for the physician to take a voice note during the examination and have it converted and structured into text (speech-to-text) by AI. This increases bedside time and reduces typing burden. But it carries two additional risks. First, transcription error: similar-sounding medical terms may be confused (e.g. “hypo-” and “hyper-” prefixes, drug names, numbers). A "15 mg" can easily be written as "50 mg". Second, media leakage: another patient's name or information may pass in the background during recording; this unintentionally leads to invasion of privacy.

So speech-to-text output must pass two layers of checks: first for transcription accuracy (especially by listening to or comparing the dose, number, and drug names to the note), then for content accuracy (is there any fabrication or spliced ​​information). The voice recording itself is also private data; Storage, deletion and access must be managed in accordance with KVKK.

Attention: The most dangerous error in the text produced from a voice memo is the silent number error. Always verify the dose and numerical values ​​separately; It is not enough that it "sounds right" to the ear.

In summary

AI saves a lot of time in writing anamnesis and epicrisis; turns the messy note into an orderly, coherent text. But the clinical document is a legal record and can add non-AI information. Always give the rule of "only use the information I give, don't add anything", compare the output line by line with the actual file, delete anything made up, and sign as a physician. Never skip anonymizing credentials.

Application task

Turn the brief notes of an anonymous case into an epicrisis with the powerful template above. Then you had the AI ​​compare its output to the original notes: how many lines were made up or added? Find and delete them manually, fill in the [MISSING] fields, and review the final text. Take note of which headline you see made-up information most often.

checklist

  • [ ] I collected the notes anonymously and completely.
  • [ ] I made clear the rule of "adding information" to AI.
  • [ ] I compared the output line by line with the actual file.
  • [ ] I deleted fake findings, values ​​and dates.
  • I filled in the [ ] [MISSING] fields with actual data.
  • [ ] I verified that there are no credentials left.
  • [ ] As a physician, I read, corrected and signed.
Previous Next